Medical Informatics
Although a number of medical decision-support systems have been developed,
little attention has been given to effective delivery of expert
information to medical personnel involved in
patient-centered activities such as trauma care. TraumAID is a
decision support system for addressing the initial definitive
management of emergency center trauma care; it was
developed by researchers at the University of Pennsylvania and
the Medical College
of Pennsylvania. In research with these colleagues and Terry Harvey,
we have been investigating how information might
be delivered so that it has the greatest positive impact on
patient care. Our analysis of critiques produced by the
critiquing module associated
with the TraumAID decision support system showed that while each
critique was coherent in isolation, the critiques as a whole exhibit
informational overlap and incoherence. To address this problem,
we developed a message planner that takes the text plans
for an arbitrary but inter-related
set of communicative goals and produces a concise and
coherent integrated message. The message planner addresses the
problems of informational overlap in the original text plans and the
appearance of conflict between text plans, and exploits relations
between text plans to enhance coherence. Although the original text planner
(TraumaGEN) was domain-dependent, our new text planner (RTPI)
utilizes domain-independent rules along with adjustable parameters
that determine when and how rules are invoked.
The second component of our medical informatics
research is concerned with exploiting the
extensive knowledge bases in decision support systems.
We constructed TraumaCASE, a system that does reverse-chaining
on the TraumAID rules and accesses knowledge stored in the
TraumAID knowledge base to
automatically construct realistic clinical cases of varying levels
of difficulty. Such cases can be used for instructional purposes by
a training module or for recertification exams by a quality assurance
module. Automatic case generation eliminates the need to collect
and pre-store a library of cases and reduces the likelihood that a
selected cases replicates one used previously.
Relevant Publications
(with B. Webber, J. R. Clarke, A. Gertner, T.
Harvey, R. Rymon, and R. Washington) Exploiting Multiple Goals
and Intentions in Decision Support for the Management of Multiple Trauma:
A Review of the TraumAID Project. Artificial Intelligence Journal,
105(1-2), pp. 263-293, 1998.
(gzipped postscript paper) x
(with T. Harvey) Integrating Text Plans for
Conciseness and Coherence. Proceedings of the 36th Annual Meeting of
the Association for Computational Linguistics and the 17th International
Conference on Computational Linguistics (ACL-COLING), pp. 512-518, 1998.
(with T. Harvey and J. R. Clarke M.D.)
Integrating Communicative Goals for Real-time Clinical Decision Support.
Proceedings of the American Medical Informatics Association
Annual Fall Symposium (AMIA), pp. 734-738, 1997.
(with T. Harvey) Generating Coherent Messages in
Real-time Decision Support: Exploiting Discourse Theory for Discourse
Practice. Proceedings of the Nineteenth Annual Conference of the
Cognitive Science Society, pp. 79-84, 1997.
(postscript paper)
(with J. R. Clarke M.D.)
TraumaCASE: Exploiting the Knowledge Base of an Existing Decision Support
System to Automatically Construct Medical Cases.
Proceedings of the Tenth International Symposium on Methodologies for
Intelligent Systems (ISMIS), pp. 456-466, 1997.
(with J. R. Clarke M.D.)
Generating Clinical Exercises of Varying Difficulty. Proceedings of the
Sixth International Conference on User Modeling (UM-97), pp. 273-275, 1997.
(with J. R. Clarke M.D. and A. Gertner)
Automatic Construction of Medical Cases for Training and Testing
Using the Knowledge Base of an Existing Decision Support System.
Proceedings of the AAAI Spring Symposium on Artificial Intelligence
in Medicine, pp. 16-20, 1996.
Last updated: Sept. 10, 1998
carberry@cis.udel.edu
Back to my homepage.